def __init__(self, model_type, filename, labels=None, feature_extractor=None, model=None):
self._update_only_on_error = None
self.model_type = model_type
self.filename = filename
if feature_extractor is not None and model is not None:
self.feature_extractor = feature_extractor
self.model = model
return
if model_type == config.SPARSE_PERCEPTRON:
from classifiers.sparse_perceptron import SparsePerceptron
from features.sparse_features import SparseFeatureExtractor
self.feature_extractor = SparseFeatureExtractor()
self.model = SparsePerceptron(filename, labels, min_update=Config().args.minupdate)
elif model_type == config.DENSE_PERCEPTRON:
from features.embedding import FeatureEmbedding
from classifiers.dense_perceptron import DensePerceptron
self.feature_extractor = self.dense_features_wrapper(FeatureEmbedding)
self.model = DensePerceptron(filename, labels, num_features=self.feature_extractor.num_features())
elif model_type == config.FEEDFORWARD_NN:
from features.enumerator import FeatureEnumerator
from nn.feedforward import FeedforwardNeuralNetwork
self.feature_extractor = self.dense_features_wrapper(FeatureEnumerator)
self.model = FeedforwardNeuralNetwork(filename, labels, feature_params=self.feature_extractor.params,
layers=Config().args.layers,
layer_dim=Config().args.layerdim,
activation=Config().args.activation,
normalize=Config().args.normalize,
init=Config().args.init,
max_num_labels=Config().args.maxlabels,
batch_size=Config().args.batchsize,
minibatch_size=Config().args.minibatchsize,
nb_epochs=Config().args.nbepochs,
dropout=Config().args.dropout,
optimizer=Config().args.optimizer,
loss=Config().args.loss,
regularizer=Config().args.regularizer,
regularization=Config().args.regularization
)
else:
raise ValueError("Invalid model type: '%s'" % model_type)
class Model(object):
def __init__(self, model_type, filename, labels=None, feature_extractor=None, model=None):
self._update_only_on_error = None
self.model_type = model_type
self.filename = filename
if feature_extractor is not None and model is not None:
self.feature_extractor = feature_extractor
self.model = model
return
if model_type == config.SPARSE_PERCEPTRON:
from classifiers.sparse_perceptron import SparsePerceptron
from features.sparse_features import SparseFeatureExtractor
self.feature_extractor = SparseFeatureExtractor()
self.model = SparsePerceptron(filename, labels, min_update=Config().args.minupdate)
elif model_type == config.DENSE_PERCEPTRON:
from features.embedding import FeatureEmbedding
from classifiers.dense_perceptron import DensePerceptron
self.feature_extractor = self.dense_features_wrapper(FeatureEmbedding)
self.model = DensePerceptron(filename, labels, num_features=self.feature_extractor.num_features())
elif model_type == config.FEEDFORWARD_NN:
from features.enumerator import FeatureEnumerator
from nn.feedforward import FeedforwardNeuralNetwork
self.feature_extractor = self.dense_features_wrapper(FeatureEnumerator)
self.model = FeedforwardNeuralNetwork(filename, labels, feature_params=self.feature_extractor.params,
layers=Config().args.layers,
layer_dim=Config().args.layerdim,
activation=Config().args.activation,
normalize=Config().args.normalize,
init=Config().args.init,
max_num_labels=Config().args.maxlabels,
batch_size=Config().args.batchsize,
minibatch_size=Config().args.minibatchsize,
nb_epochs=Config().args.nbepochs,
dropout=Config().args.dropout,
optimizer=Config().args.optimizer,
loss=Config().args.loss,
regularizer=Config().args.regularizer,
regularization=Config().args.regularization
)
else:
raise ValueError("Invalid model type: '%s'" % model_type)
@staticmethod
def dense_features_wrapper(wrapper):
from features.dense_features import DenseFeatureExtractor
params = [
FeatureParameters("w", Config().args.wordvectors, Config().args.maxwords, Config().args.worddropout),
FeatureParameters("t", Config().args.tagdim, Config().args.maxtags),
FeatureParameters("e", Config().args.labeldim, Config().args.maxedgelabels),
FeatureParameters("p", Config().args.punctdim, Config().args.maxpuncts),
FeatureParameters("x", Config().args.gapdim, Config().args.maxgaps),
FeatureParameters("A", Config().args.actiondim, Config().args.maxactions),
]
return wrapper(DenseFeatureExtractor(), params)
def extract_features(self, *args, **kwargs):
return self.feature_extractor.extract_features(*args, **kwargs)
def score(self, *args, **kwargs):
return self.model.score(*args, **kwargs)
def update(self, *args, **kwargs):
self.model.update(*args, **kwargs)
def finish(self, train):
self.model.finish(train)
def advance(self):
self.model.advance()
@property
def update_only_on_error(self):
if self._update_only_on_error is None:
self._update_only_on_error = self.model_type in (config.SPARSE_PERCEPTRON, config.DENSE_PERCEPTRON)
return self._update_only_on_error
def finalize(self, *args, **kwargs):
return Model(model_type=self.model_type,
filename=self.filename,
feature_extractor=self.feature_extractor.finalize(*args, **kwargs),
model=self.model.finalize(*args, **kwargs))
def save(self):
if self.filename is not None:
try:
self.feature_extractor.save(self.filename)
self.model.save()
except Exception as e:
raise IOError("Failed saving model to '%s'" % self.filename, e)
def load(self):
if self.filename is not None:
try:
self.feature_extractor.load(self.filename)
self.model.load()
Actions().all = self.model.labels
except Exception as e:
raise IOError("Failed loading model from '%s'" % self.filename, e)
